Minimal Gifu Anaerobic Broth Media (mGAM) (Hyserve 1,005,433) was prepared per manufacturing instructions; briefly, 41.7 g was suspended per liter of H₂O and heated to allow to dissolve before autoclaving at 121°C for 15 min. All mGAM was supplemented with 1 mg/mL of mucin (Sigma-Aldrich M1778) prior to autoclaving to allow for dissolving. Media was placed in anaerobic chamber at least 24 h in advance to experimentation to allow for exchange of gasses. Desulfovibrio Postgate Medium (DPM) was prepared using a modified version of a DSMZ established protocol (DSMZ, n.d.); briefly, Solution A (0.5 g/L K₂HPO₄, 1.0 g/L NH₄Cl, 1.0 g Na₂SO4, 0.1 g CaCl₂ × 2H₂O, 2.0 g MgSO₄ × 7H₂O, 2.0 g Na-L-lactate, 1.0 g Yeast extract, 980 mL H₂O), Solution B (0.5 g FeSO₄ × 7H₂O, 10 mL H₂O), and Solution C (0.1 g Sodium thioglycolate, 0.1 g Ascorbic acid, 10 mL H₂O), were prepared while spinning. Solution A was filter sterilized using a 0.22 μm bottle vacuum filter (Corning 430,015) using sterile flame technique followed by autoclaving at 121°C for 15 min, and Solutions B and C were filter sterilized using 0.22 μm centrifuge tube filters (Millipore SCGP00525) by sterile flame technique. Solutions B and C were added to room temperature Solution A and immediately placed in the anaerobic chamber to reduce. Working media for culturing (10%DPM/mGAM+mucin) was created by diluting a ratio of 1 mL DPM in 9 mL of mGAM + 1 mg/mL mucin using sterile flame techniques.

A Coy Lab Products vinyl anaerobic chamber (Coy Labs 7,150,000) was used to create anaerobic conditions for all cultures and culturing conditions. The vinyl anaerobic chamber is outfitted with an incubator (Coy Labs 6,100,000) set at 37°C, a recirculating HEPA atmospheric filtration system (Coy Labs 8,537,025 V), an anaerobic gas infuser (Coy Labs 8,100,110) and a dehumidifier (8,533,110) to ensure anaerobic conditions. The chamber is also outfitted with an anaerobic monitor, model 12 (Coy Labs 6,250,000) to allow for monitoring of anaerobic conditions and hydrogen % content. Atmospheric makeup of the machine is created using a mixture of two gasses: pure nitrogen gas, and a mixed gas containing 5% CO₂, 5% H₂, and balance N₂. Gasses are mixed into the chamber to obtain a 2.0–2.5% hydrogen content, and an O₂ = 0–10 ppm.

Eight bacterial species were chosen to create a synthetic community: Akkermansia muciniphila, Bacteroides fragilis, Bacteroides thetaiotaomicron, Bifidobacterium longum, Desulfovibrio desulfuricans, Escherichia coli, Lactobacillus johnsonii, and Roseburia hominis. Supplementary Table S1 includes specific vendor information for bacterial strains purchased. All species, except for D. desulfuricans and B. longum, were grown in 5 mL mGAM +1 mg/mL mucin media for 24 h, with the two exceptions being grown in 5 mL DPM for 48 h. Then 150 μL of each culture was inoculated into a plate and diluted with equal parts sterile filtered PBS. OD₆₀₀ measurements were taken using the Molecular Devices SpectraMax M3 Spectrophotometer using SoftMax Pro v6, blanking the culture to its original medium. Cultures were then consolidated in the following ratios in 10% DPM/mGAM + 1 mg/mL mucin to achieve an OD₆₀₀ of 0.1: 1 part each B. thetaiotaomicron, B. fragilis, E. coli, and B. longum to 5 parts each R. hominis, L. johnsonii, D. desulfuricans, and B. longum.

Cecums from male C57BL/6 J mice (Charles River) were collected per IACUC protocol 23-05-0013. Mice were then dissected, and the entire intestinal tract was collected into a sterile plate and transferred to the anaerobic chamber. In the anaerobic chamber, the cecum was isolated from the rest of the GI tract and its contents (approximately 200 μL) were transferred into approximately 15 mL of 10% DPM/mGAM + 1 mg/mL mucin media and cultured for approximately 2 h.

A human stool sample was obtained through a collaboration between Brown University and Hasbro Children’s Hospital (IRB 1930822-3) by Rahiya Rehman, and immediately frozen at −80°C. Approximately 130 mg of stool was weighed out and placed into 20 mL of 10% DPM/mGAM + 1 mg/mL mucin, pipetted to break up the matter, and grown for 24 h while shaking at ~400 rpm. Cultured content was then frozen at −80°C in 1 mL aliquots in 20% filtered glycerol (Fisher Sci G331) until the start of the experiment. On the day of the experiment, frozen cultures were diluted 1:200 in media prior to the start of the experiment (this sample is used for Supplementary Figure S1 only).

For cultures treated with iron and amoxicillin, a human stool sample was obtained through BioIVT (BioIVT HUMANFECES-0000263); donor information was deidentified prior to receipt of material. Upon arrival, the sample was immediately frozen at −80°C. Approximately 130 mg of stool samples were weighed out and placed into 20 mL of 10% DPM/mGAM + 1 mg/mL mucin, pipetted to break up matter, and grown until exponential growth phase (~0.2) for 5 h while shaking at ~400 rpm. Content was then frozen at −80° C in 1 mL aliquots in 20% filtered glycerol (Fisher Sci G331) until the start of the experiment. On the day of the experiment, frozen culture content—originating from the single donor—is diluted 1:200 in media and allowed to grow until once again at the exponential growth phase of approximately 0.2 (~6 h) before treatment (this sample is used for experimental data shown in Figures 2, 3).

Input cultures, prior to treatment, were first grown to an optical density of approximately 0.2. Cultures were then treated with hemin (Fisher Sci AAA1116503), FeSO₄ (Sigma-Aldrich F8633) or FeCl₃ (Fisher Sci AC169430050), and then 300 μL was added per well to a 96 well flat bottom plate (Corning 3,370) and treated with amoxicillin (Millipore A8523) at concentrations of 50 μg/mL, 10 μg/mL, 12.5 μg/mL, 3.125 μg/mL, 2 μg/mL, or 0 μg/mL using amoxicillin stocks diluted in DMSO (Fisher Sci BP2311). Cultures’ optical densities were monitored for 24 h using the Cerillo Alto Kinetic Microplate Reader while incubating at 37°C in the anaerobic chamber, and cultures were collected at 24 h post treatment for downstream applications.

Cultures collected post treatment were processed via a selected portion of the Zymo Research HostZERO Microbial DNA kit (Zymo D4310). Briefly, 300 μL of sample was centrifuged at 2,250 × g for 8 min to pellet solids, then the supernatant was removed. One microliter of Microbial Selection Enzyme (Zymo D4310350) in 100 uL of Microbial Selection Buffer (Zymo D431025) was added to the pellet, vortexed, and incubated at 37°C for 15 min. Samples were then treated with 20 uL of Proteinase K (D30012125, reconstituted at 20 mg/mL), vortexed, incubated at 55°C for 10 min, then diluted 1:1 in DNA/RNA Shield (R1100250, 1X Solution) and frozen at −80°C until further use.

E. coli Nissle 1917 (see Supplementary Table S1 for more information) was grown to stationary phase in mGAM + 1 mg/mL mucin (OD ~ 3–4) then diluted to an OD₆₀₀ of 0.5. These E. coli cultures were then treated for 30 min with amoxicillin (Millipore A8523) at a concentration of 100 μg/mL and then mixed 1:1 into the synthetic community previously described, also mixed to a final OD₆₀₀ of 0.5. These spiked communities were then treated with or without the DNA Cleanup step and had their 16S rRNA gene amplified and sequenced as described below.

DNA for all samples was extracted using the ZymoBIOMICS Quick-DNA Fecal/Soil Microbe 96 Kit (Zymo D8011) following the manufacturer’s protocol. Total DNA was eluted into nuclease-free water and quantified using the dsDNA-BR kit on a Qubit 4.0 fluorometer (ThermoFisher Scientific, Waltham, MA, United States) before library preparation.

16S rRNA V4-V9 hypervariable regions were amplified from total DNA using barcoded 515F forward primers from the Earth Microbiome Project (Thompson et al., 2017) and using a matched set of reverse barcodes alongside the 1492R primer. See Supplementary Table S2 for reverse barcoded primer sequences used. Amplicons were created using 5X Phusion HF DNA Polymerase (Fisher Sci F530L) under the following PCR conditions: an initial denaturation at 98° C for 30 s followed by 25 cycles of 98°C for 10s, 57°C for 30 s, and 72°C for 30 s. Amplicons then underwent a final extension at 72°C for 5 min. Amplicons were visualized for confirmation of successful amplification via gel electrophoresis, and pooled at equal DNA weight before PCR cleanup using the Macherey-Nagel Gel and PCR Clean-Up kit (Macherey-Nagel 740609.250) following the manufacturer’s protocol. DNA libraries were prepared with Oxford Nanopore Technologies’ Native Barcoding 24 v14 kit (ONT SQK-NBD114-96), then sequenced with a MinION MK1B (ONT MIN-101B).

16S rRNA sequences were basecalled using Nanopore’s Guppy Software v6.5.7 through the guppy_basecaller function. Reads were then demultiplexed using Nanopore’s Guppy Software v6.5.7 using the guppy_barcoder function, and barcodes and adapters were trimmed. Primary demultiplexed reads were then secondarily demultiplexed using Porechop with custom barcode inputs denoted from the Earth Microbiome Project (Thompson et al., 2017) and Supplementary Table S2. Taxonomic assignments were performed using the SILVA database (Quast et al., 2013) with Emu (Curry et al., 2022). Shannon Diversity index values and Bray–Curtis dissimilarity values were calculated using the phyloseq package v1.46.0 (McMurdie and Holmes, 2013) in RStudio v12.1 (McMurdie and Holmes, 2013; Lozupone and Knight, 2005).

Bacterial species denoted in Supplementary Table S1 were grown anaerobically in mGAM + 1 mg/mL mucin for 24 h prior to the experiment. Experimental 96-well flat bottom plates (Corning 3,370) were setup with dilutions of amoxicillin (Millipore A8523) ranging in concentration from 0 to 250 μg/mL (final concentrations). Bacterial cultures were then diluted 1:1000 into media containing no iron, or media with supplementations of different concentrations of hemin (Fisher Sci AAA1116503), FeSO₄ (Sigma-Aldrich F8633) or FeCl₃ (Fisher Sci AC169430050), and 150 uL was then inoculated per well into the plates. Plates were then incubated at 37°C for 20 h before diluting 1:1 with 1X PBS (Fisher Sci BP39920) and OD₆₀₀ measurements were taken using the Molecular Devices SpectraMax M3 Spectrophotometer using SoftMax Pro v6. Percent growth was then calculated against 0 μg/mL treated cultures, and the Gompertz Equation for MIC Determination was used to determine MIC₉₀ with Prism GraphPad v10.4.1.

Specific details pertaining to statistical analyses for all experiments are detailed in the figure legends and results section. MaAsLin2 (Mallick et al., 2021) was used to analyze outputs from Emu. T-tests and Gompertz Equation for MIC Determination were performed in Prism GraphPad v10.4.1. All other graphs were created using Prism GraphPad v10.4.1.

Though ex vivo culturing approaches have been used to identify changes in microbiomes (Hirano et al., 2023; Tao et al., 2023; Cheng et al., 2022; Goodman et al., 2011), large passage numbers of the communities meant to stabilize the microbiomes cause the passaged communities to differ substantially from original community inputs (Aranda-Díaz et al., 2022; Celis et al., 2023); a protocol is thus necessary for cultured communities to better resemble the original communities in question. A brief overview of the protocol developed is seen in Figure 1A. Communities, ranging from known synthetic communities to complex unknown human stool microbiome samples, were cultured out and treated with amoxicillin, with their optical density monitored. The cultured samples then have their 16S rRNA V4-V9 region amplified using a double barcoding approach modified from the Earth Microbiome Project (Thompson et al., 2017) and sequenced on the Nanopore MinION, then demultiplexed and aligned using the Emu algorithm (Curry et al., 2022). The Emu taxonomic alignment algorithm more accurately discerns species taxonomic information and overcome any potential barcode leakage previously seen with Nanopore sequencing (Bemis et al., 2025).

To test the validity of this pipeline, we cultured a synthetic community containing eight microbial species combined to act as the initial community input for the pipeline: A. muciniphila, B. fragilis, B. thetaiotaomicron, B. longum, D. desulfuricans, E. coli, L. johnsonii, and R. hominis. These microbes were chosen to represent a wide range of phyla, and more information on these species can be found in Supplementary Table S1. When this community was treated with 10 μg/mL amoxicillin for 24 h, there was a noticed increase in the relative abundance of E. coli in the community (Figure 1B). Because of the treatment of this contained community for 24 h with a bactericidal drug, it was possible that DNA from dead microbes, including E. coli could be represented downstream in the taxonomic abundances. To reduce the impact of DNA from killed microbes, we chose to use a DNA cleanup step derived from Zymo Research’s Host-Zero Microbial DNA Kit, comprising of a DNase treatment followed by a Proteinase K treatment prior to DNA extraction (Heravi et al., 2020). To test this step, we spiked our synthetic community with an equal OD₆₀₀ of E. coli killed with 100 μg/mL of amoxicillin for 1 h and treated with or without the DNA cleanup steps to eliminate free floating DNA (Figure 1C). Treating the community with the DNA Cleanup alone caused a limited shift in the relative abundance of the community and a general reduction in the relative abundance of E. coli. When killed E. coli was spiked into a base community containing viable E. coli, the relative abundance of E. coli dominated as expected. Performing a cleanup on this “spike-in” community reduced the abundance of E. coli to mirror the community that had not been spiked with killed E. coli, indicating that the proposed cleanup was effective at eliminating killed bacteria.

While this cleanup step worked well for a known synthetic community, we still needed to confirm its effectiveness for mammalian samples. Additionally, we needed a way to compare relative abundance with the overall growth of the culture. In order to accomplish this, murine cecal samples were collected as the complex community input detailed in Figure 1A and treated with different amoxicillin concentrations for 24 h followed by the presence/absence of our DNA Cleanup step prior to amplification and sequencing (Figure 1D). Data generated from the pipeline was normalized to OD₆₀₀ to integrate overall community growth with the relative abundance data. Treatment of the communities with amoxicillin created a stepwise reduction in total normalized relative abundance of the murine cecal samples. The implementation of the DNA cleanup step showed changes in a few families of bacteria within the communities. Families in bold reduce in normalized relative abundance as a result of the DNA cleanup, indicating that these families are overrepresented and experience more drastic killing as a result of the culturing and antibiotic treatment; these families identified include Butyricicoccaceae and Enterobacteriaceae. Families underlined indicate an increase in normalized relative abundance as a result of the DNA cleanup, indicating that these families are normally underrepresented without it; these families include Acholeplasmataceae, Akkermansiaceae, Bacteroidaceae, Desulfovibrionaceae, and Lactobacillaceae.

The pipeline was also tested using a patient stool sample that was cultured and frozen in 20% glycerol to act as starting communities in the pipeline. These samples retained their Shannon diversity over a 48 h culturing period and showed minimal variance and dissimilarity through Bray-Curtis beta-diversity when compared to the donor sample (Supplementary Figures S1A,B). Profiling of these samples in Supplementary Figure S1C showed that, although there were some shifts in families over time such Bifidobacteriaceae, Sutterellaceae, and Tannerellaceae, and Muribaculaceae, these shifts were not able to alter overall alpha and beta diversity form our donor sample. In order to identify species retention over time of these samples, a “Percent of Species Still Present from Baseline” value was calculated (Supplementary Figure S1D), where the percentage of species that retained at least 1% of their relative abundance values compared to the donor stool sample was identified. Culturing of the stool sample for 24 h allowed for a 1% species retention threshold of about 77%, while culturing the communities further out to 48 h resulted in a 1% species retention threshold of 43.25%. As populations reach their maximum OD600 over the course of the 48 h, the number of viable bacteria growing in culture was likely reduced due to toxicity of stationary phase cultures. This data particularly highlights the effectiveness of the culturing portion of the pipeline to retain species over time; it should be noted that increasing total culturing times begins to cause a degradation in the total number of species represented, so special care should be taken to reduce total culturing time.

Iron is a critical factor in bacterial metabolism, and previous research has shown that it may influence antibiotic efficacy (Ezraty and Barras, 2016; Cuisiniere et al., 2021). Given that human-derived components are a major source of iron in mGAM media, we selected hemin as a potential modulator of antibiotic lethality. Since iron can serve as a terminal electron acceptor in metabolic processes and anaerobic respiration (Nealson and Saffarini, 1994; Richter et al., 2012), we hypothesized that altering hemin concentrations within the media could push microbial communities toward a more energy-productive state, thereby enhancing antibiotic lethality. To test this hypothesis, we used a human microbiome biospecimen obtained from BioIVT as the input community.

We found that hemin supplementation alone slightly reduced the initial growth rate of the microbiome culture but ultimately resulted in a higher final OD₆₀₀, seen in Figure 2A. When combined with 2 μg/mL or 10 μg/mL amoxicillin, hemin had a modest effect on early culture inhibition kinetics but led to higher final ODs compared to antibiotic treatment alone. This suggests that hemin facilitates microbiome recovery following initial AMX inhibition, potentially aiding post-antibiotic microbiome restoration or supporting bacterial survival in the presence of antibiotics.

We sequenced this community and first profiled alpha diversity at 24 h, finding that AMX led to the expected significant drop in Shannon diversity in both the + 0 μM and + 15 μM hemin groups (^#p < 0.05). On the other hand, the + 60 μM hemin cultures started with a significantly lower Shannon diversity (*p < 0.05) without amoxicillin treatment compared to the + 0 μM hemin group and exhibited no significant changes in Shannon diversity as a result of amoxicillin treatments (Figure 2B). The Bray–Curtis dissimilarity beta diversity plot of the samples showed distinct concentration-driven separation along PC2 for amoxicillin either at zero or 15 μM hemin (Figure 2C). The presence of 60 μM hemin shifted the community into the positive PC1 quadrant. The low amoxicillin concentration was also shifted into that quadrant, while the high AMX concentration returned the hemin community to the zero-iron AMX-treated group.

We next examined the OD-normalized relative abundance of the communities and observed shifts in four major genera in response to the combined effects of increased hemin and amoxicillin: Bacteroides, Escherichia-Shigella, Parabacteroides, and Parasutterella, shown in bold in the Legend (Figure 2D). The significance of these changes was confirmed through MaAsLin2, and independently graphed and shown in Figures 2E–G and Supplementary Figure S2A. The Bacteroides genus, without additional hemin, initially bloom in low amoxicillin but are eventually reduced at higher amoxicillin concentrations. The addition of + 15 μM hemin changes this dynamic, stabilizing the relative abundance numbers for the group when treated with antibiotics. When given + 60 μM hemin, Bacteroides are further stabilized, and even bloom with AMX (Figures 2D,E). We tested the MIC of two representative species of this genus, B fragilis and B thetaiotaomicron. The addition of increasing amounts of hemin to culture media did not significantly shift the MIC₉₀ of either of these species (Supplementary Figures S2C,D). It should be noted that the Bacteroides genus shifting in Figure 3D is comprised almost exclusively of an undetermined species of Bacteroides, and this phenomenon originally observed may be unique to this species.

In the presence of amoxicillin, Escherichia-Shigella does not expand until supplemented with + 60 μM hemin, allowing for a shift in overall resistance to amoxicillin and allowing for major growth in 2 μg/mL of amoxicillin. High hemin concentrations also support a large bloom in normalized relative abundance of Escherichia-Shigella without the presence of AMX (Figures 2D,F). We also found with E. coli Nissle 1917 that hemin supplementation increased the MIC₉₀ at least 10-fold (Figure 2H). It should be noted that this strain of E. coli is not the same strain seen in our dataset. Parabacteroides and Parasutterella both exhibited a bloom in the presence of amoxicillin and high hemin concentrations in the media (Figures 2D,G; Supplementary Figure S2A) indicating that hemin may have a protective effect for these genera against amoxicillin.

The genus Enterococcus showed no significant change to amoxicillin susceptibility in response to increasing amoxicillin but did show interesting nonsignificant trends, leading us to the impact of hemin on the MIC₉₀ of an E. faecalis strain. We found that giving this strain hemin at any concentrations led to a drastic drop in the MIC₉₀ to about 1/10 of control conditions (Supplementary Figures S2B,E), indicating that hemin drastically sensitized this strain of E. faecalis to antibiotic killing.

Iron can exist in many forms, both trapped in molecules like hemin (chloroprotoporphyrinIX iron(III)), or in soluble forms in both ferric (Fe³⁺) and ferrous (Fe²⁺) iron. To test whether the phenomena identified with hemin was conserved among iron species or unique compared to free iron sources, the communities were grown and monitored for 24 h in the presence of either Fe(II)SO₄ or Fe(III)Cl₃. The presence of free iron, either ferrous (II) or ferric (III), did not affect overall optical density of the cultures (Figure 3A). This holds true between amoxicillin treated groups, where it is shown that amoxicillin causes a stepwise reduction in OD₆₀₀ but this drop is not influenced by the presence of free iron. The 24 h Shannon diversity showed a drop in Shannon diversity in response to amoxicillin treatment in control iron groups, and this trend does not change between introduction of different sources and concentrations of iron, except the introduction of + 75 μM FeCl₃, which causes a reduction in the base Shannon diversity compared to the control iron group without amoxicillin, and a significant drop (^#p < 0.05) in diversity when 2 μg/mL amoxicillin is added (Figure 3B). The Bray–Curtis dissimilarity PCoA plot showed that samples clustered by presence or absence of amoxicillin on the PC1 axis, and no difference in clustering between iron concentrations or groups (Figure 3C). The normalized relative abundance breakdown of the treated communities can be seen in Figure 3D, with the Bacteroides, Escherichia-Shigella, Enterococcus, and Parabacteroides genera shown separately (Figures 3E,F; Supplementary Figures S3A,B).

Looking at the OD normalized relative abundance we found that in the absence of added free iron, Bacteroides was elevated by 2 μg/mL AMX but was fully eliminated at 10 μg/mL AMX. Iron supplementation in both free forms led to an elimination of Bacteroides at the protected low AMX concertation. This is a direct contradiction to the protection we observed with hemin, where further increases in hemin lead to increased amoxicillin protection. Percent growth curves generated for B. fragilis and B. thetaiotaomicron show that there is no change in MIC₉₀ values for either species when grown in different concentrations and types of free iron (Supplementary Figures S3C,D), indicating that iron in these strains does not increase amoxicillin susceptibility in the presence of different free iron sources. It should again be noted that the species identified within these samples contains mostly an unidentified species, and the unique effect hemin has as well as the lack of protection for free iron may be specific to this strain.

Similarly, Escherichia-Shigella did not show the same level of protection with free iron (Figures 3D,F) as with hemin (Figures 2D,F). We did observe that a higher concentration of FeCl₃ limits the growth of Escherichia-Shigella when grown without amoxicillin (Figure 3F), though amoxicillin susceptibility did not significantly change because of different forms of iron. The MIC₉₀ of E. coli Nissle however does not change when given any type or concentration of iron (Supplementary Figure S3E). This contradicts what was seen with hemin, where additions of hemin increased the MIC₉₀ at least 10-fold. This indicates that E. coli is particularly adapted to using hemin to resist amoxicillin killing and not matching concentrations of free ferric or ferrous iron.

Parabacteroides do not grow well with the addition of any free form of iron, regardless of amoxicillin introduction (Supplementary Figure S3A), which is contrary to the result seen with hemin in Figure 3D and Supplementary Figure S3G where the bacteria was protected by supplementation and bloomed with hemin. The Enterococcus genus was also isolated and monitored. The addition of amoxicillin in media with no iron supplementation cause a stepwise reduction of Enterococcus (Supplementary Figure S3B). The genus also experienced a slight bloom in response to a small amount of amoxicillin when given FeSO₄ but not when given FeCl₃ and experienced a drop afterwards with large amounts of amoxicillin. The MIC₉₀ for E. faecalis showed no changes in response to the introduction of different iron supplementations (Supplementary Figure S3F). This contradicts the results seen in Supplementary Figure S1F, where supplementation of hemin immediately potentiated the susceptibility of amoxicillin. This strain is not the dominant strain or species identified in our dataset, and it should be noted that this may be unique to E. faecalis UWH 1921.